Spaces:
Sleeping
Sleeping
| import cv2 | |
| import numpy as np | |
| from mtcnn import MTCNN | |
| from keras_facenet import FaceNet | |
| class FaceVerification: | |
| """ | |
| A class for performing face verification using MTCNN for detection | |
| and FaceNet for embedding extraction. | |
| Attributes | |
| ---------- | |
| detector : MTCNN | |
| Face detector model. | |
| embedder : FaceNet | |
| Face embedding model. | |
| Methods | |
| ------- | |
| extract_face(img, target_size=(160, 160)): | |
| Detects and crops the face from the input image. | |
| get_embedding(face): | |
| Extracts L2-normalized FaceNet embedding from the given face image. | |
| cosine_similarity(a, b): | |
| Computes cosine similarity between two embedding vectors. | |
| compare_faces(img1, img2, threshold=0.5): | |
| Compares two face images and determines if they belong to the same person. | |
| """ | |
| def __init__(self): | |
| """Initialize the MTCNN detector and FaceNet embedder.""" | |
| self.detector = MTCNN() | |
| self.embedder = FaceNet() | |
| print("✅ MTCNN and FaceNet initialized successfully.") | |
| def extract_face(self, img, target_size=(160, 160)): | |
| """ | |
| Detects and crops a face from an image. | |
| Parameters | |
| ---------- | |
| img : np.ndarray | |
| Input image in BGR format. | |
| target_size : tuple, optional | |
| Desired size for the cropped face (default is (160, 160)). | |
| Returns | |
| ------- | |
| np.ndarray | |
| Cropped and resized face image in RGB format. | |
| Raises | |
| ------ | |
| ValueError | |
| If no face is detected or if the image is invalid. | |
| """ | |
| if img is None: | |
| raise ValueError("Image not found or invalid input.") | |
| img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) | |
| detections = self.detector.detect_faces(img_rgb) | |
| if not detections: | |
| raise ValueError("❌ No face detected.") | |
| # Use first detected face | |
| x, y, w, h = detections[0]['box'] | |
| x, y = abs(x), abs(y) | |
| face = img_rgb[y:y + h, x:x + w] | |
| # Resize to FaceNet input size | |
| face = cv2.resize(face, target_size) | |
| return face | |
| def get_embedding(self, face): | |
| """ | |
| Extracts L2-normalized FaceNet embedding. | |
| Parameters | |
| ---------- | |
| face : np.ndarray | |
| Cropped face image (RGB format, 160x160). | |
| Returns | |
| ------- | |
| np.ndarray | |
| L2-normalized face embedding vector. | |
| """ | |
| emb = self.embedder.embeddings([face]) | |
| emb = emb / np.linalg.norm(emb, axis=1, keepdims=True) | |
| return emb[0] | |
| def cosine_similarity(self, a, b): | |
| """ | |
| Computes cosine similarity between two embeddings. | |
| Parameters | |
| ---------- | |
| a, b : np.ndarray | |
| Face embeddings. | |
| Returns | |
| ------- | |
| float | |
| Cosine similarity score. | |
| """ | |
| return np.dot(a, b) | |
| def compare_faces(self, img1, img2, threshold=0.5): | |
| """ | |
| Compares two face images and prints similarity result. | |
| Parameters | |
| ---------- | |
| img1, img2 : np.ndarray | |
| Two input images (BGR format). | |
| threshold : float, optional | |
| Similarity threshold (default is 0.5). | |
| Returns | |
| ------- | |
| json | |
| of keys STATUS, ERROR (None if no error) | |
| """ | |
| try: | |
| face1 = self.extract_face(img1) | |
| face2 = self.extract_face(img2) | |
| except Exception as e: | |
| return {"STATUS":"Failed, try again","ERROR":str(e)} | |
| emb1 = self.get_embedding(face1) | |
| emb2 = self.get_embedding(face2) | |
| similarity = self.cosine_similarity(emb1, emb2) | |
| accuracy = ((similarity + 1) / 2) * 100 | |
| print(f"Cosine similarity: {similarity:.4f}") | |
| if similarity > threshold: | |
| print("✅ Same person (accept)") | |
| return {"STATUS":f"✅ Same person (accept), accuracy {accuracy:.2f}%", "ERROR":"None"} | |
| else: | |
| return {"STATUS":f"❌ Different persons (reject), accuracy {accuracy:.2f}%", "ERROR":"None"} | |